Add support for gpt-4-turbo

[simonw]

Announced at OpenAI DevDay today - ~3x cheaper than GPT-4, and has a 128,000 token limit.

[simonw]

Not working yet (I guessed the model name):

llm --model gpt-4-turbo 'hi'                                 

Error: The model `gpt-4-turbo` does not exist

[simonw]

Now working in this branch: https://github.com/simonw/llm/compare/gpt-4-turbo

[simonw]

Oh it's gpt-4-1106-preview according to https://openai.com/blog/new-models-and-developer-products-announced-at-devday

Not working yet though:

$ llm -m gpt-4-turbo 'hi'
Error: The model `gpt-4-1106-preview` does not exist or you do not have access to it.
Learn more: https://help.openai.com/en/articles/7102672-how-can-i-access-gpt-4.

https://help.openai.com/en/articles/7102672-how-can-i-access-gpt-4.

[fourseven]

Appreciate the work you've done with LLM, it's a very useful tool. I followed the Adding more OpenAI Models instructions for the new model to set it up, and wanted to post this for others until it's officially released/supported.

For what it's worth, my account also doesn't yet have access, the following results in Error: The model 'GPT4 Turbo' does not exist

Edit: I had a human readable model name instead of the alias in the below config, have updated it now and I get data back!

llm -m gpt-4-1106-preview "Generate 5 names for a pet mice that are puns"

1. Squeakheart - A pun on "sweetheart" with a nod to the characteristic mouse squeak.
2. Mice-ter Whiskers - A play on "Mister Whiskers" emphasizing the "mice" part.
3. Cheese Chaser - A humorous take on the stereotype of mice loving cheese.
4. Whisker Twister - A fun blend of "whisker" and "twister" to reflect a mouse's nimble movements.
5. Gouda Buddy - A cheesy twist on "good buddy," referencing a well-known type of cheese.

# $(llm logs path)/extra-openai-models.yaml

- model_id: gpt-4-1106-preview
  model_name: gpt-4-1106-preview
  aliases: ['gpt-4-turbo', '4-turbo', '4t']

[simonw]

Demo to summarize comments on https://news.ycombinator.com/item?id=38166420

curl -s "https://hn.algolia.com/api/v1/items/38166420" | \
  jq -r 'recurse(.children[]) | .author + ": " + .text' | \
  llm -m gpt-4-turbo 'Summarize the themes of the opinions expressed here,
  including direct quotes in quote markers (with author attribution) for each theme.
  Fix HTML entities. Output markdown. Go long.'

Output: https://gist.github.com/simonw/d50c8634320d339bd88f0ef17dea0a03

[simonw]

Appreciate the work you've done with LLM, it's a very useful tool. I followed the Adding more OpenAI Models instructions for the new model to set it up, and wanted to post this for others until it's officially released/supported.

Nice! I am glad that feature worked as intended.

[cmungall]

For anyone following the instructions here: https://llm.datasette.io/en/stable/other-models.html#adding-more-openai-models

I had to add both model_id and model_name e.g.

- model_name: gpt-4-1106-preview
  model_id: gpt-4 turbo preview
  aliases: ["4-turbo", "gpt-4-turbo"]

[shvechikov]

What about adding an explicit support for JSON mode introduced in GPT-4 Turbo?

openai.Completion.create(
    ...
    response_format={ "type": "json_object" }
)